Hands-on prompt engineering and LLM evaluation. Built around a diagnose-then-ship loop: forensic transcript critique ▎ that names the exact failure mode, paired with a fixer that produces complete, deployment-ready rewrites — not ▎ vague advice.
A full analysis pipeline that turns a dataset into a finished report, not just charts. Hand it a CSV, Excel, Parquet, or JSON file and it cleans the data, scores its quality, runs EDA, and where the data supports it, trains a prediction model with SHAP attribution and a backtested forecast. You get a trust score, ranked findings, and recommendations as a Word doc, an interactive HTML report, and an optional PowerPoint deck. The numbers come from code, so they stay consistent, never invented.
Compresses AI filler text — preambles, transitions, repetitions — at 3 intensity levels (Lite / Full / Ultra) while enforcing a protected scope (URLs, numbers, code, reasoning) that is never deleted, so "be concise" stops silently destroying what matters.
I specialize in safe Codex skill onboarding. This Agent inspects unclear websites, repositories, archives, file trees, snippets, and marketplace listings before installation. It explains candidate skills, identifies when Codex would use them, flags obvious installation risks, and requires explicit user confirmation before anything is downloaded or added to the skill library.
Four-layer AI detection evasion, not word swapping. Real math (burstiness CV, vocabulary density) — no guessing. Detector-specific: attacks GPTZero, Originality, Turnitin by their weakest links. Chinese NLP: tracks 四字成语密度, a blue ocean. 105+ AI markers with genre-aware replacements. Research-backed (27 reports, 25+ papers). Honest: 75-85% typical improvement, no fake 100% claims.
Create a reversible, project-local Codex copy of Claude Code agents, skills, commands, compatible hooks, MCP, and persistent GPT routing—while keeping Claude untouched.
Tired of the 30-second cap on AI video generation and expensive video APIs? Tell your agent what you sell, or paste your website. Promo Studio writes the script, shows it to you, then renders finished MP4s with natural voiceover and word-by-word captions. One request gives you TikTok, YouTube and square versions, test hooks, subtitles and thumbnails. Everything runs on your computer, so there's no cost per video.
Turn open-ended survey responses into a reviewable codebook, coded CSV, theme summary, quote table, and insight deck outline. 将开放式问卷回答整理成可人工复核的 codebook、编码表、主题摘要、引用表和洞察汇报大纲。No fabrication, no PII leakage, no statistical overclaiming.
更少 Token，更精準的 AI 指令。以語意最佳化移除冗餘、合併重複要求，保留真正意圖、必要限制、輸出格式與重要上下文；提供 Token 估算，並可依要求比較優化前後差異。支援 MAX_COMPRESSION、BALANCED、MAX_PRECISION。